We ship the AI layer with the code
Every repo we hand back carries instruction files, specs, skills, a DESIGN.md, and an MCP server. What each one is for, and how our people and the agents split the work of building them. When we finish a project, the client keeps the code. That used to be the whole list. Now the layer that drafted the code has to come with it, or the client's own team cannot extend the system the way we did. This…
<The AI layer is included with the code in every repository handed back to the client.>
Every project delivered contains essential instruction files, specifications, and a DESIGN.md file, along with an MCP server. The contents of the repository provide the client with the necessary knowledge to continue building the system as we originally designed it.
<The formats used to convey information to agents and engineers include AGENTS.md, OpenSpec, and DESIGN.md.>
AGENTS.md outlines how the system is constructed, the pull request gates, data restrictions, and the location of runbooks. It serves as the primary reference for both agents and new engineers, ensuring consistency across the project.
<Requirements are maintained in the repo using OpenSpec, an open-source specification-driven development tool compatible with various coding agents.>
Two folders, openspec/specs and openspec/changes, store the current system behavior and proposed changes separately. As changes are implemented, they are archived with a date, creating a historical record for future reference.
<Skills are folder-based instructions that agents load when specific tasks are required.>
Most common tasks already have pre-existing skills, including database migrations, routing structures, and quality scoring rubrics. However, custom skills are written when no suitable existing solution is available. These skills are maintained alongside the code and change through the same pull request process.
<The DESIGN.md file provides the design system to agents in a format they can utilize.>
This markdown file contains tokens with real values, type ramps, spacing habits, component rules, and the reasoning behind them. By including a DESIGN.md file alongside the code, clients can ensure the client's AI-generated components align with the established design system.
<An MCP server is provided to enable non-engineers to interact with the system using conversational agents.>
The MCP server allows clients to query the system through a conversational agent like Claude or ChatGPT, retrieving answers in real-time based on the analyzed and scored data. This significantly reduces the time required for manual analysis and provides instant access to information.
<The process of creating drafts involves the collaboration of multiple team members.>
In each stage of the development cycle, agents assist by drafting initial requirements, variations, and edge cases. However, the final review and approval are still necessary, as agents may generate results with predictable inaccuracies. The drafts are reviewed through a pull request process, ensuring the quality and accuracy of the final deliverables.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
